Re: chrooting daemons

1997-10-16 Thread Ben Pfaff
[EMAIL PROTECTED] (Bruce Perens) writes: > We don't want to have more than one instance of a library in the > filesystem if that is at all possible. We currently violate this for > anonymous FTP, which has its own copy of a few system utilities and > its own copy of the C shared library. I notice t

Re: chrooting daemons

1997-10-16 Thread Kai Henningsen
[EMAIL PROTECTED] (Bruce Perens) wrote on 16.10.97 in <[EMAIL PROTECTED]>: > The best way to deal with this would be some sort of filesystem hack under > the chroot-ed directory. What would be needed would be a way to read-only access specified parts of the outside filesystem, preferrably with

Re: chrooting daemons

1997-10-16 Thread Bruce Perens
From: Topi Miettinen <[EMAIL PROTECTED]> > If we had a fexec(int fd,..) [...] > I don't know how dynamic libs would be dealt with, though. Not easily. What I want to do is to have a list that is made into a filesystem in the chroot environment, but uses the same device/inode pairs as the underlyi

Re: chrooting daemons

1997-10-16 Thread Bruce Perens
From: Topi Miettinen <[EMAIL PROTECTED]> > But if the libraries are not hard linked (that probably would be ruined in > update), I believe they would still be in memory more than once, because > they then had different inodes and the OS would have no way of knowing they > are the same. We don't wa

Re: chrooting daemons

1997-10-16 Thread Bruce Perens
From: [EMAIL PROTECTED] (Richard Braakman) > It might be even easier to link the daemon statically. No, the small security or filesystem integrity gain of using a static-linked binary is not worth the cost of having a second copy of libraries in its working set. The developers have discussed this

Re: chrooting daemons

1997-10-16 Thread joost witteveen
> It might be even easier to link the daemon statically. It's not going to > be able to share its libraries with any other program anyway. Linking them statically may open other security holes: there are bugs found in the libraries. Fixing the libc bugs will then also require the system admin to

Re: chrooting daemons

1997-10-16 Thread Richard Braakman
It might be even easier to link the daemon statically. It's not going to be able to share its libraries with any other program anyway. Of course, that means that sites that want that kind of security will have to compile the daemons themselves. I expect that they'll want to do that anyway. If I

Re: chrooting daemons

1997-10-16 Thread Bruce Perens
Before you run a shared-library program in a chroot context, you need to provide the shared libraries in that context. It's easier to have the daemon chroot itself after it's already been loaded. It's also nice to have the daemon drop root privileges, but it's not always possible. Bruce --

chrooting daemons

1997-10-15 Thread Topi Miettinen
One way to limit damage caused by a daemon with an exploited security hole is to run it in "jail", chrooted environment. An example is DNS: http://www.homeport.org/~adam/dns.html The same mechanism could be used with several other daemons which do not need to access files outside their running dir